Towards Interactive Robots in Autism Therapy
Towards Interactive Robots in Autism Therapy
Towards Interactive Robots in Autism Therapy
<LINK "dau-n*">
"dau-r76">
"dau-r20">
"dau-r52">
"art" TITLE "Articles">
<TARGET "dau" DOCINFO AUTHOR "Kerstin Dautenhahn and Iain Werry"TITLE "Towards interactive robots in autism therapy"SUBJECT "P&C, Volume 12:1"KEYWORDS ""SIZE HEIGHT "220"WIDTH "150"VOFFSET "4">
1. Introduction
1960’s, Papert and his colleagues at the MIT AI LAB developed the program-
ming language LOGO that has been widely used in teaching children. A remote
controlled device (a ‘turtle’ robot) was developed which moved according to a
set of LOGO instructions (cf. the LEGO/LOGO Artificial Life Toolkit for
children; Resnick 1989). The LOGO programming language and LEGO robots
have since been used widely in education.
It is expected that a new generation of children will increasingly use
computer technology in a variety of contexts (professional, educational and
recreational), including interactive robotic toys, digitally enhanced objects and
tangible interfaces (Laurel 1993; Tapscott 1998; Cassell and Jenkins 1999; Druin
and Hendler 2000). New interactive systems and novel interfaces are also likely
to impact on methods of therapy and rehabilitation. ‘Persuasive’ technology
(Fogg 1999) is technology that can influence the opinions, attitudes and
behaviour of people. In particular, the physical shape and behaviour of socially
intelligent agents, that display aspects of human-style social intelligence
(Dautenhahn 1998; Dautenhahn (ed) 2000), are likely to change how we can
teach social intelligence to humans who have difficulties in understanding and
displaying social behaviour.
Recently, Socially Intelligent Agents research has resulted in a variety of
different software and robotic systems which can successfully interact with
humans and show aspects of human-style social intelligence (for an overview
see Dautenhahn and Numaoka 1998, 1999; Dautenhahn (ed) 2000; SIA 2000;
Dautenhahn et al. (eds) 2002). Interesting interactive robotic systems include
the Kismet platform (Breazeal and Scassellati 1999) and the Robota dolls
(Billard et al. 1998; Billard 2000). Kismet is a humanoid face that can generate
expressive social interactions with human ‘caretakers’. Such ‘meaningful’
interactions can be regarded as a stepping-stone for the development of social
relationships between a robot and a human. The Robota dolls are humanoid
robots, developed as interactive toys for children, used as research platforms in
order to study how a human can teach a robot, using imitation, speech and
gestures. Increasingly, robotic platforms are being developed to be interactive
playmates for children (e.g., Montemayor et al. 2000; Cañamero and Fredslund
2000). Besides commercial purposes (see Sony’s Aibo robot), such interactive
robotic systems can potentially be utilised as learning environments or in
rehabilitation/therapy applications, as studied in the Aurora project, which we
describe in more detail in the next section.
<LINK "dau-r64">
"dau-r49">
"dau-r27">
2.1 Autism
Autism is a developmental disorder defined by diagnostic criteria specified in
DSM-IV (Diagnostic and Statistical Manual of Mental Disorders, American
Psychiatric Association, 1994). Individuals with autism show a broad spectrum
of difficulties and (dis)abilities, and they vary greatly in their levels of overall
intellectual functioning. However, all individuals diagnosed with autism posses
a common set of symptoms. The National Autistic Society (NAS 2003) lists the
following triad of impairments:
1. Social interaction (difficulty with social relationships; for example, appear-
ing aloof and indifferent to other people, inappropriate social interactions,
inability to relate to others in a meaningful way, impaired capacity to
understand others’ feelings or mental states).
2. Social communication (difficulty with verbal and non-verbal communica-
tion; for example, not really understanding the meaning of gestures, facial
expressions or tone of voice).
3. Imagination (difficulty in the development of play and imagination; for
example, having a limited range of imaginative activities, possibly copied
and pursued rigidly and repetitively).
In addition to this triad, repetitive behaviour patterns and a resistance to change
in routine can generally be observed. These are associated with a significantly
reduced repertoire of activities and interests, stereotypical behaviour, and a
tendency of fixation to stable environments. Rates of occurrence are given
which range between 5–15 in 10000. Unlike a physical handicap, which pre-
vents people from physically interacting with the environment, people with
autism have great difficulty in making sense of the world, in particular the social
world. Autism can, but need not, be accompanied by learning disabilities. For
example, at the higher functioning end of the autistic spectrum we find people
with Asperger Syndrome. Some of them manage to live independently as adults
and to succeed in their profession, but only by learning and applying explicit
rules in order to overcome the ‘social barrier’ (cf. the autobiographic accounts
by Grandin (1995), Grandin and Scariano (1996), Schäfer (1997)). Instead of
picking up and interpreting social cues ‘naturally’ they must learn and memor-
ise rules about what kind of behaviour is socially appropriate during interaction
with non-autistic people. Autism is not, as is often assumed, a voluntary
<LINK "dau-r43">
"dau-r75">
decision to retract from the world: people with autism do not have the choice
to live socially or not; the decision has been made for them.
Two different viewpoints exist about how best to support people with
autism: either it is argued that efforts should be undertaken to teach people with
autism the skills they need to interact and survive in the world of typically-
developed people, or it is suggested that they might be happier living separately
in a world specifically designed for them. Empowering people with autism,
allowing them to make their own choice on whether or not to link with the
world of non-autistic people poses many challenges. In order to understand
people with autism we have to not only better understand the causes of autism,
but also find ways to empower them, for example by using computer and
robotic technology, so that they have the choice of whether and to what extent
they want to connect to other people.
1. a
1. b
Figure 1.(a) The toy truck (left) and the mobile robot Labo-1 (right) used in the
comparative study. Labo-1’s basic sensor configuration consists of active infrared
sensors for obstacle avoidance and pyro-electric sensors, which allow detection and
following of humans. The robot weighs about 6.5 kg. Due to a four-wheel differential
drive it can turn very smoothly. The robot can manage a few kilograms of additional
weight, e.g. when children are pushing the robot or (partially) stepping on it. Words and
simple phrases are produced by the robot in certain situations, by means of a voice
production device (not shown). In the trials the robot moves very slowly, so that even
when it bumps into a child (which rarely happens since the children are very attentive
to the robot’s movements) no harm is done. The robot is robust enough to cope with
being extensively pushed around during the trials.
(b) Two children with autism interacting with the Labo-1 robot. The photo on the left
shows a child who is playing a ‘chasing game’ with the robot. The photo on the right
shows a child who played with the robot for an extended period of time until he needed
to go back to class. He was not afraid to interact with the robot at very close distance.
Most of the time the child was lying on the floor and playing ‘interaction games’ with
the robot, i.e., reaching out and touching the robot which then caused the robot to
approach/avoid the child.
<LINK "dau-r59">
"dau-r34">
"dau-r20">
Figure 1 shows one of the mobile robots used in the Aurora project. In this
paper we concentrate on trials we conducted with this particular robot. The
children who were interacting with the robot were between 8–12 years of age,
and included non-verbal children, i.e., children who could not use language or
usually did not use language.
Teaching methods for children with autism that address therapeutic issues
(e.g., eye contact, joint attention, turn taking, reading mental states, and
emotions) usually involve constrained teaching sessions (e.g., Howlin et al. 1999).
In contrast, robot-human interactions in the Aurora project are deliberately
chosen to be playful, unconstrained and unstructured. Specifically, the children
are allowed to interact with the robot in whatever position they prefer (e.g.,
lying on the floor, crawling, standing — cf. Figure 1). They are also free to
choose how they interact with the robot (touching, approaching, watching from
a distance, picking it up, etc.). Interference by adults is only necessary when the
child is about to damage the robot, or when the child (by pressing buttons)
switches off the robot so that it needs to be restarted. These conditions are very
different from those typical of other projects on robot-human interaction (e.g.,
Kismet or the Robota dolls, both mentioned above), where the human is
expected to interact with the robot while adopting a particular position and
orientation towards the robot (e.g., sitting face-to-face in close vicinity of an
interactive robot that is not moving).
2.4.1 Mindreading
Generally humans are, from an early age on, attracted to self-propelled objects
that move autonomously and seemingly with ‘intention’ (Dautenhahn 1997).
Premack and Premack (1995) present a theory of human social competence that
consists of three units: the first unit (intentional system) identifies self-propelled
movements in space and interprets them as produced by intentional objects that
are engaged in goal-directed behaviour, such as escaping from confinement,
making contact with another intentional object, overcoming gravity (e.g.,
seeking to climb a hill). In this way animate and inanimate objects can be
distinguished, since only animate objects can move both in space and time
without the influence of other objects (e.g., a stone can roll downhill, but not
uphill without interference by external forces). Movement in the same location,
however, is interpreted as animate but not intentional. The second unit in
Premack and Premack’s theory is the social system, which specifies the changes
that the intentional objects undergo. This allows one to interpret relations as
possession or group membership. The third unit is the theory of mind system,
which outputs explanations of the actions performed in terms of states of mind
such as perception, desire, and belief.
The effects of the ‘intentional stance’ produced by the above mentioned
mechanisms, in particular the intentional system as the basic unit which selects
the objects to be considered, were convincingly demonstrated in a classical
<LINK "dau-r44">
"dau-r32">
"dau-r51">
"dau-r58">
"dau-r46">
"dau-r14">
"dau-r1">
experimental study (Heider and Simmel 1944). In this study, human subjects
created elaborate narratives about intentional agents when asked to describe
movements of animated geometric shapes shown in a silent film. Based on
Heider and Simmel’s studies, more recent studies have confirmed and further
refined their results. For example, Oatley and Yuill (1985) showed that cues
signaling a social context increase people’s tendency to use personal and
mentalising descriptions. Rimé et al. (1985) demonstrated that the replacement
of geometric shapes by human-like shapes reduced the intensity with which
subjects attribute emotional states, thus showing that patterns of movement are
more powerful in evoking anthropomorphic interpretations than the appear-
ance of the characters. It has been shown that children as young as three years
can interpret simple patterns of movements as intentional, goal-directed
behaviour (Montgomery and Montgomery 1999). Thus, the interpretation of
patterns of movements as ‘goal-directed’ and ‘intentional’, and the interpreta-
tion of interactions among such moving objects as ‘social’ or ‘mentalistic’, is
fundamental to social intelligence in pypically developing humans.
Bowler and Thommen (2000) showed that children with autism are able to
distinguish mechanical motion from intentional action as well as control groups
of children who were matched according to chronological and verbal mental
age. Abell et al. (2000) investigated the attribution of mental states in children
with autism in more detail. Using computer animations of geometrical objects,
similar to the original Heider and Simmel study, they distinguished between
interpretations in terms of actions in descriptions of randomly moving objects
(e.g., an object bouncing off a wall), interpretation of goal-directed (G-D)
behaviour in descriptions of objects that interacted with each other (e.g.,
fighting, following each other, etc.), and the use of mentalising descriptions for
sequences that involved characters responding to ‘mental states’ (e.g., mocking,
persuading, tricking, etc.). In the random condition the subjects were told to see
‘just triangles’, in the G-D condition the objects were given animal roles, and in
the theory-of-mind condition the objects were identified as people. Results
show that children with autism have impairment in using appropriate mental
concepts when describing animations that involved mental states, even when
these same children passed standard false belief tasks. This demonstrates the
need to study mentalising capabilities for interpreting real-time interactions
that might not be adequately captured by standard false belief tasks.
A more general behaviour reading mechanism is also suggested as the basis
for anthropomorphism (Mitchell and Hamm 1997): evidence indicates that, for
evoking anthropomorphic interpretations using narratives the behaviour of
<LINK "dau-r62">
"dau-r13">
"dau-r11">
"dau-r20">
"dau-r25">
"dau-r41">
"dau-r6">
objects (in Mitchell and Hamm’s study animals were used) is more important
than other aspects, e.g., the appearance of an object or whether a human is
familiar with the object. People seem to apply these behaviour-reading mecha-
nisms quite readily to inanimate objects such as robots. Every robotics researcher
who has ever given a demonstration of autonomous mobile robots to a general
audience can confirm how readily humans view robots as people (Bumby and
Dautenhahn 1999).
There are many theories that aim to explain autism at a biological, develop-
mental, cognitive, or neurobiological level. For example, the narrative deficit
hypothesis of autism suggests a vital role of preverbal transactional processes in
early development (Bruner and Feldman 1993; see also discussion in Dauten-
hahn 2002). However, the hypothesis of a specific theory of mind deficit in
autism is at present the most widely accepted theory. Many researchers have
contributed to the ToMM framework, including Uta Frith (see Frith et al.
1991), Alan M. Leslie (1994), and Simon Baron-Cohen (see Baron-Cohen et al.
1985). It should be noted, however, that a deficit in theory of mind cannot
account for the whole spectrum of symptoms that are characteristic of autism.
For example, a deficit in executive function (e.g., in planning, impulse control,
initiation of action, etc.) has been proposed that could account for the tendency
towards repetitive and stereotypical behaviour (Russell 1997).
Premack and Premack’s theory of human social competence bears similari-
ty to Baron-Cohen’s suggestion (1995) of four mechanisms underlying the
human mindreading system. The first is the intentionality detector (ID) that
interprets motion stimuli (stimuli with self-propulsion and direction) in terms
of the mental states of goal and desire. These mental states are basic since they
allow for making sense of universal movements of all animals, namely approach
and avoidance, independent of the form or shape of the animal. The ID
mechanism works through vision, touch and audition and interprets anything
that moves with self-propelled motion or producing a non-random sound as an
object with goals and desires. The second mechanism in Baron-Cohen’s
mindreading system is the eye-direction detector (EDD), which only works
through vision. EDD detects the presence of eye-like stimuli, detects the
direction of the eyes, and interprets gazing as seeing (attribution of perceptual
states). This mechanism allows for the interpretation of stimuli in terms of what
an agent sees. ID and EDD represent dyadic relations (relations between two
objects, agent and object, or agent and self), such as ‘Agent X wants Y’ or
‘Agent X sees Y’; however, they do not allow for the establishment of a link
between what another agent sees and wants and what the self sees and wants.
<LINK "dau-r1">
interaction of robots with children with autism. Indeed, our work focuses
primarily on the social role of imitation (Dautenhahn 1994), i.e., imitation as a
format of communication that creates intersubjectivity in human-human
interaction. We propose that the important link between the dynamics of
imitation, turn-taking and social interactions can be explored with robots.
From an Artificial Intelligence perspective, we previously suggested a
conceptual framework to classify different and increasingly complex dynamics
in robot-human interactions (Dautenhahn and Werry 2000). Within this
framework, robot-human interactions in the Aurora project are designed where
synchronisation of movements, temporal coordination, and potentially the
emergence of imitation games can be used as important mechanisms for ‘social
contact’ between the robot and the child. It is hoped that such an approach,
focusing on interaction dynamics rather than cognitive reasoning mechanisms,
can incrementally facilitate and strengthen the temporal aspects fundamental to
the development of social competence and the ability to socially interact with
people (cf. Hall 1983; Farnell 1999).
There are potentially at least two different ways that contingent and
imitative behaviour can be used in the Aurora project. Firstly, can a robot that
shows behaviour contingent with a child’s performance or that imitates a child’s
movements increase the child’s social behaviour (cf. Escalona et al. 2002)?
Secondly, can we encourage a child to imitate a robot’s movements? In the
latter case, either the robot might be used to assess a child’s imitation skills, or
the imitative behaviour might be used as a stepping-stone towards more
complex interaction. In our work with the mobile robot we focussed on very
simple imitation games where the robot and the child take turns, e.g., in
approach and avoidance games. The ‘imitative behaviour’ that we study is
therefore far simpler than the conditions usually used in autism research.
Note that ‘social behaviour’ in the context of robot-child interaction is
quite different from child-adult or child-child interaction. It is therefore not
clear whether any ‘social behaviour’ shown by a child towards a robot will be
generalised and applied to interactions with people (see below, for a discussion
on generalisation).
able and less complex than human behaviour. However, the robot’s behaviour
is not completely predictable. For example, the same ‘approach child’ behaviour
will never be repeated precisely; it will rather be performed in variations. This
issue is important to our work, since we must avoid perpetuating stereotypical
and repetitive behaviour that is characteristic of autism.
Initial trials in the Aurora project stressed the individual nature of the
specific needs of children with autism, but they also showed that most children
responded very well and with great interest to the autonomous robot. In a series
of comparative trials, where the children played with a mobile robot (condition 1)
and also (separately) with a passive non-robotic toy (condition 2), children showed
greater interest in interactions with the robot than with the ‘inanimate’ toy.
The study just mentioned allowed us to compare the way in which the
children interacted with the robotic platform to the way they interacted with a
standard toy. In order to evaluate robot-human interactions we developed a
80
Contact-Robot
Percentages
60 Contact-Toy
40 Eye gaze-Robot
Eye gaze-Toy
20
0
A B C D
Contact-Robot 69.03 41.95 83.9 19.24
Contact-Toy 56.91 60.55 2.99 57
Eye gaze-Robot 81.64 60.56 93.33 53.99
Eye gaze-Toy 40.24 71.67 14.18 60.33
Child
Figure 2.This figure displays selected results that are part of a larger comparative study
that involved 18 children and tested the two conditions (robot versus toy truck). Results
are shown for four boys with autism with regard to the microbehaviours eye gaze and
contact time (cf. Werry et al. 2001a). Contact time measures any instances when the
child touches, operates or handles the robot or the toy truck. Depicted are the percent-
ages of the micro-behaviours observed during the duration of the trial (approximately
four minutes of exposure to the robot followed by exposure to the toy truck). Full
results of the comparative study are presented in Werry (2003).
<LINK "dau-r61">
"dau-r20">
"dau-r70">
"dau-r76">
"dau-r39">
"dau-r40">
"dau-r78">
"dau-r67">
demonstrated that the children with autism who showed least eye gaze toward
a correct location (associated to the task) had the most severe autistic character-
istics. Many training programmes for children with autism include the training
of eye gaze behaviours and eye contact. The training criteria for ‘normal eye
gaze frequencies’ do however vary tremendously. As shown by Arnold et al.
(2000), peer-to-peer eye gaze in small playgroups of familiar typically develop-
ing children (5–10 years old) were significantly less than previously reported
data that was based on adult-child or adult-adult dyads. Thus, it remains
unclear how much ‘eye gaze’ behaviour should be taught to children with
autism. But whatever the ‘normal eye gaze frequencies’ are, research suggests
that eye gaze is an educationally and therapeutically relevant behaviour. In our
work we assume that it can be used as an indicator of interest- and engagement-
level, similar to attention and contact behaviours, although we are aware that
for children with autism the correlation between these behaviours and levels of
engagement/interest might be less straightforward than for typically developing
children. Any interpretation of our data therefore must be cautious.
For more details on the quantitative and qualitative evaluations of children
with autism interacting with robots in the Aurora project, see Werry et al.
(20001a,b), Dautenhahn and Werry (2002), Dautenhahn et al. (2002b), Werry
(2003). Evaluation is a major issue in our work. A range of different qualitative
as well as quantitative evaluation techniques are needed in order to reveal not
only statistical regularities and patterns, but also meaningful events of behaviour in
context. For instance, an application of Conversation Analysis has revealed interest-
ing aspects of how the robot can elicit communication and interaction competen-
cies of children with autism (Dautenhahn et al. 2002; Robins et al., in press).
2.4.4 An example
In order to exemplify the types of interaction that occur in the Aurora project,
we now provide a narrative account of one particular boy with autism interact-
ing with a mobile robot, displaying some ‘typical’ behaviour also observed in
other children with autism. The example shows different ways of how a boy
with autism plays with a mobile robot. This indicates that, in addition to the
quantitative observational data used to assess the occurrence of micro-behav-
iours such as eye gaze (cf. Figure 2), other behaviours are difficult to evaluate
without considering the context in which they occur. We shall call the boy with
autism that took part in the comparative study Olaf.
Olaf interacted with the robot before he was exposed to the toy truck. On
entering the room, he investigated the toy that was located on the floor and
Towards interactive robots in autism therapy 19
then he exited the room. When he re-entered the room, the trial started. On
entering the room for the second time, he looked around before one of the
investigators showed him the robot and turned it on. As this was happening
Olaf again looked around the room. There seemed to be not enough to keep his
attention, as the robot was not acting at that point. However, twelve seconds
into the trial, he again noticed the robot and approached it. The robot was still
inactive and he walked away from it again. However, he continuously glanced
at the robot, for periods between one and three seconds. Thirty seconds into the
trial, the robot began to move as one of the investigators prompted it to move.
This seemed to catch Olaf ’s attention: he again looked at the robot and stopped
moving around the room. He continued to gaze at it for eight seconds, before
moving away. However, he continued to watch the robot as he moved around
it for thirteen seconds. At this point, he became interested in the experimenter
and the video camera. After moving around the room, he avoided the robot as
it moved towards him, but kept observing it. Seventy-five seconds into the trial,
Olaf bent down to interact with the robot at its level, looking around it, investigat-
ing it and touching it gently. When the investigator turned the robot around and
it headed towards him, Olaf was not disturbed by the robot’s movement towards
him, and did not attempt to move away, even though he was seated on the
floor. He even leaned lower to gaze at the sensor of the robot, commonly
treated by the children as the ‘front’ of the robot. In total, Olaf was seated on
the floor for seventy seconds, and spent fifty seven seconds gazing at the robot.
Two minutes and forty seconds into the trial, Olaf reached for the toy truck and
brought it out, interacting with both the toy and the robot simultaneously.
Initially, Olaf was more interested in the toy than the robot, but twenty
seven seconds after the toy was brought out, he used the toy to interact with the
robot. He did this by placing the toy in front of the robot, and it backed away.
It is significant to note that Olaf did not engage in this type of operation of the
robot without the toy being there, and that no one had told him that the robot
would behave in this way. He then continued to ‘push’ the robotic platform
backwards by placing the toy in front of it, so as to trigger the robot’s forward
sensors. Eighty seconds after the toy was brought out, Olaf switched his
attention back to the robot, ignoring the toy and gazing at the robot for twenty
seven seconds, at which point Olaf again distributed his attention between the
toy and the robot, operating the robot by using the toy.
This type of behaviour is typical for Olaf. He appeared cautious in his
interactions, and looked around the room frequently. He kept a distance from
the robot and appeared uncertain about interaction of any kind. He seemed to
<LINK "dau-r63">
"dau-r76">
"dau-r59">
"dau-r20">
be encouraged to interact with the robot through using the toy rather than in a
direct manner. This might indicate a willingness to engage in interaction. It is
hoped that, with further familiarity with the robot, Olaf will be more confident
in the way that he uses the robot, to a point where he might be able to explore
the interactions and to learn from them.
Several other trials have been conducted in the Aurora project. For exam-
ple, in a different set of trials (dual-child trials) we investigated three pairs of
children with autism who were simultaneously exposed to the robot. Different
play styles could be observed. These trials point towards the possible role of the
robot as a mediator — a device that mediates interactions between people
(Werry et al. 2001b; Robins et al. 2004; Robins et al., in press). This mediating
role is very different from how we used the robot in the single child trials, where
the robot’s purpose was to teach basic interaction skills through play. Generally,
besides the obvious role of a robot as a therapeutic interaction partner, different
roles of robots in autism therapy can be envisaged (Dautenhahn 2003).
preferences, and that can guide the child through different therapeutically
relevant interactions. This is by far our most long-term goal.
In this section we outline some of the design issues that play an important role
in the Aurora project. We then discuss these in the context of interactive virtual
environments, in order to demonstrate differences or commonalities between
robotic and virtual environments in autism therapy.
children with autism to explore their own bodies and how they interact with
their environment. Thus, bodily interaction itself can be as therapeutically
relevant as the ‘content’ of the interaction.
3.4 Generalisation
A major problem in all therapeutic approaches to autism is generalisation: A
child often shows improved performance in the particular teaching environ-
ment (e.g., in classroom) but has great difficulty in generalising the learning
experience and applying the newly acquired skill to non-classroom situations.
Virtual environments have a good potential in this respect: Creating different
contexts and environments in the classroom and changing features and shapes
of objects in the environment is very time-consuming and often not feasible,
whereas creating alternative scenarios or variations in virtual environments is
comparatively easy. This is important for specific learning objectives as well as
for a broader approach, e.g., the general facilitation of imaginative skills. To give
an example: If a teacher enacts a story together with children, then the colour of
a blanket cannot be changed instantaneously, nor can a sword suddenly appear
out of thin air. Typically developing children can easily compensate for these
‘deficiencies’ of the real world, for their imaginative skills allow them to create
different worlds and alternative or fictional realities, as shown by role-play.
However, as the imaginative skills of children with autism are often impaired,
they prefer the concrete, the visible. The shape of a robot cannot change
suddenly; it cannot grow wings and fly away. However, in virtual environments
rich, dynamic, and at the same time concrete and visible worlds can be created,
although mostly limited to the visual (a child doesn’t get wet if it starts raining,
the feeling of raindrops on the skin cannot be experienced ‘virtually’). However,
the issue of how a child with autism can transfer skills learnt in virtual environ-
ments to the real world remains unsolved.
3.5 Presence
‘Presence’ is a key issue in designing virtual environments (e.g., Heeter 1992):
The acceptance of virtual environments does depend strongly on whether the
user’s presence in the artificial environment is believable, i.e., whether he or she
has the impression of ‘being there’. Often reality is confusing to a person with
autism; clear boundaries, meaning, and order seem to be missing. Thus, for
children with autism the feeling of ‘being there’, in the real world, is likely to be
<LINK "dau-r55">
"dau-r25">
"dau-r30">
on features and details of objects and sensations in their environment that are
of little interest to other people. For example, if an object is presented to the
child one cannot assume that the child directs his/her attention to the object as
a whole; it is rather likely that the child’s perception will focus on aspects of the
object, e.g., colour, shape, texture, structural details, etc. Persistent preoccupa-
tion with parts of objects is also used as a diagnostic criterion in DSM-IV. In the
Aurora project it is therefore vital to ‘hide’ any wires, buttons or other features
of the robot that might attract the children’s attention. Such changes in the
environment are easier to achieve in a virtual environment where different
aspects of the world can be hidden, highlighted, dynamically changed (depend-
ing on the child’s activities), presented differently, etc.
4. Conclusion
This article introduced the project Aurora and discussed its background,
motivation as well as particular challenges and issues involved in developing
interactive robotic systems as therapeutic teaching devices for children with
autism. Some comparisons were drawn between robotic and virtual learning
environments. It is hoped that the development of robust and believable
interactive systems (robotic and software) can support the education and
therapy of children with autism, so that ultimately such technologies can
become an integrated part of the curriculum, being used by teachers and
parents and tailored towards specific individual needs of children with autism.
Providing an enjoyable and entertaining ‘toy’ specifically adapted to the needs
of the children, thereby increasing the quality of life of children with autism, is
an integral part of the Aurora project. However, helping children with autism
develop social skills is methodologically and technically demanding. Given the
nature of autism, only long-term studies will reveal if and how this goal can be met.
From an Artificial Intelligence research point of view children with autism
constitute a very special user group with immense heterogeneity. It is unlikely
that one specific robotic design can be used generically. Based on our experi-
ence thus far we assume that only a range of robotic designs can meet the
particular requirements of particular groups of children and individuals. Thus,
the design space of interactive environments needs to be explored and linked to
the space of sets of requirements, the niche space (Sloman 1995). One might
speculate that (different types of) robotic therapeutic tools might serve for
(different sets of) requirements addressing primarily bodily, physical interaction,
<LINK "dau-r54">
"dau-r9">
while (different types of) virtual environments might serve for (different sets of)
requirements addressing primarily imaginative and cognitive skills. There might
be niches for various types of interactive environments with virtual or robotic
agents which could be used in the therapy of children with autism, e.g., human-
oid and non-humanoid robots, multi-media interactive environments, digitally
enhanced toys, and virtual environments ranging from desktop VE’s to immer-
sive interactive learning and play environments (Bobick et al. 1999; Penny
2000). Among the big challenges is the development of appropriate design
methodologies and evaluation methods, so that different interactive environ-
ments and their effectiveness in the application domain of autism therapy and
education can be assessed and compared.
Future robot developments and trials with children with autism will tell
whether our initial goal, namely to facilitate social interaction skills of children
with autism in interactions with robots, can be reached. It needs to be demon-
strated whether and how the children can generalise any skills learnt during play
with the robot to peers or adults. Note that the development of an ‘attachment’
of a child with autism to a robot is not our primary aim; we rather hope to
further the interaction skills of a child with autism with people. Future results
will clarify to what extent robots can contribute to either educational or
therapeutic aims.
One issue in the Aurora project which is at present quite unclear, and which
we have therefore not yet addressed explicitly, is the role of affective aspects in
child-robot interactions. We exploit the fact that children with autism seem to
find interactions with a mobile robot rewarding and enjoyable (we observe
many positive facial and vocal expressions during the trials), but we do not
address the affective dimension of interaction specifically. Yet, if we succeed in
teaching a child to show pro-active social behaviour skills towards a robot, can
this be achieved without any affective feedback from the interaction partner, as
it is the case with robots? Can we teach imitative skills with a machine, i.e.,
without genuine intersubjectivity? How would ‘intersubjectivity’ between a
child with autism and a robot look like? Should one ‘fake’ or ‘simulate’ affect in
robots, e.g., by equipping robots with articulated faces and other means to
express emotions? If we do, what are the ethical issues involved in encouraging
the development of affective attachment of a child with autism with a robot that
is not more than a machine, and that does not possess genuine emotions? From
the perspective of a person with autism, and her needs, are these ethical
concerns really relevant?
<LINK "dau-r20">
<DEST "dau-n*">
"dau-r10">
"dau-r1">
"dau-r2">
"dau-r3">
"dau-r4">
"dau-r5">
"dau-r6">
"dau-r7">
"dau-r8">
"dau-r9">
Notes
*Takashi Gomi, director of Applied AI Systems Inc., kindly donated the robot used in the
Aurora project for the trials mentioned in this paper. We are grateful for Takashi Gomi’s
support and encouragement throughout the project. We would like to thank Patricia Beevers
and the teaching staff at Radlett Lodge School: the trials would not have been possible
without their collaboration. Many thanks to Prof. Stuart Powell, Dr. William Harwin and
other members of the Aurora team who have contributed valuable comments over the years.
Parts of this paper are based on preliminary discussions in (Dautenhahn, 2000c). The
authors would like to thank anonymous reviewers for very constructive comments on an
earlier version of this paper.
1. Among the many possible meanings of ‘remedial’, we use the term to reflect our goal of
improving (social) skills in children with autism.
References
Abell, F., Happe, F., and Frith, U. 2000. “Do triangles play tricks? Attribution of mental states to
animated shapes in normal and abnormal development”. Cognitive Development 15: 1–6.
Aibo 2000. Retrieved June 6, 2003, from http://www.aibo.com/.
Arkin, R. C. 1998. Behavior-Based Robotics. Cambridge, MA: The MIT Press.
Arnold, A., Semple, R. J., Beale, I., and Fletcher-Flinn, C. M. 2000. “Eye contact in children’s
social interactions: What is normal behaviour?” Journal of Intellectual and Developmen-
tal Disability 25(3): 207–216.
Aurora 2000. Retrieved March 19, 2004, from http://www.aurora-project.com.
Baron-Cohen, S. 1995. Mindblindness: An Essay on Autism and Theory of Mind. Cambridge,
MA: The MIT Press.
Baron-Cohen, S., Leslie, A., and Frith, U. 1985. “Does the autistic child have a ‘theory of
mind’?” Cognition 21: 37–46.
Billard, A. 2000. “Play, dreams and imitation in Robota”. Proc. AAAI Fall Symposium Socially
Intelligent Agents — The Human in the Loop. AAAI Technical Report FS-00-04, AAAI
Press, 9–12.
Billard, A., Dautenhahn, K., and Hayes, G. 1998. “Experiments on human-robot communi-
cation with Robota, an imitative learning and communicating doll robot”. Proc. Socially
Situated Intelligence Workshop. Zürich, August 1998.
Blocher, K. H. 1999. “Affective Social Quest (ASQ): Teaching emotion recognition with
interactive media and wireless expressive toys”. MSc Thesis, MIT.
Bobick, A. F., Intille, S. S., Davis, J. W., Baird, F., Pinhanez, C. S., Campbell, L. W., Ivanov,
Y. A., Schütte A., and Wilson, A. 1999. “The KidsRoom: A perceptually-based interactive
and immersive story environment”. Presence: Teleoperators and Virtual Environments
8(4): 369–393.
Breazeal, C. and Scassellati, B. 1999. “How to build robots that make friends and influence
people”. Proc. IROS99, Kyonjiu, Korea.
<DEST "dau-r20">
"dau-r11">
"dau-r12">
"dau-r13">
"dau-r14">
"dau-r15">
"dau-r16">
"dau-r17">
"dau-r18">
"dau-r19">
Bruner, J. and Feldman, C. 1993. ‘Theories of mind and the problem of autism”. In S. Baron-
Cohen et al. (eds), Understanding other Minds: Perspectives from Autism. Oxford: Oxford
University Press.
Bullowa, M. (ed). 1979. Before Speech. Cambridge: Cambridge University Press.
Bumby, K. and Dautenhahn, K. 1999. “Investigating children’s attitudes towards robots: A
case study”. Proc. CT99, The Third International Cognitive Technology Conference.
August, San Francisco, 391–410. Retrieved March 19, 2004, from http://www.cogtech.-
org/CT99/Bumby.htm.
Bowler, D. and Thommen, E. 2000. “Attribution of mechanical and social causality to
animated displays by children with autism”. Autism 4: 147–171.
Cañamero, D. and Fredslund, J. 2000. “How does it feel? Emotional interaction with a
humanoid LEGO robot”. Proc. AAAI Fall Symposium Socially Intelligent Agents — The
Human in the Loop. AAAI Technical Report FS-00-04, AAAI Press, 23–28.
Cassell, J. and Jenkins, H. (eds). 1999. From Barbie to Mortal Kombat: Gender and Computer
Games. Cambridge, MA: The MIT Press.
Charman, T. and Baron-Cohen, S. 1994. “Another look at imitation in autism”. Development
and Psychopathology 1: 1–4.
Cobb, S. V. G., Neale, H. R., and Reynolds, H. 1998. “Evaluation of virtual learning environ-
ments”. Proc. 2nd Euro. Conf. Disability, Virtual Reality and Assoc. Techn. Skövde,
Sweden, 17–23.
Cooper, M., Keating, D., Harwin, W., and Dautenhahn, K. 1999. “Robots in the classroom
— Tools for accessible education”. In C. Bühler and H. Knops (eds), Proc. AAATE
Conference 1999, The 5th European Conference for the Advancement of Assistive Technology,
November, Düsseldorf: IOS Press, 448–452.
Dautenhahn, K. 1994. “Trying to Imitate — A step towards releasing robots from social
isolation”. In P. Gaussier and J-D. Nicoud (eds), Proc. From Perception to Action Conference.
Lausanne, September. IEEE Computer Society Press, 290–301.
Dautenhahn, K. 1997. “I could be you — the phenomenological dimension of social under-
standing”. Cybernetics and Systems Journal 28(5): 417–453.
Dautenhahn, K. 1998. “The art of designing socially intelligent agents — Science, fiction, and
the human in the loop”. Applied Artificial Intelligence 12(7–8): 573–617.
Dautenhahn, K. 1999. “Robots as social actors: Aurora and the case of autism”. Proc. CT99,
The Third International Cognitive Technology Conference. San Francisco, 359–374.
Retrieved June 6, 2003, from http://www.cogtech.org/CT99/Dautenhahn.htm.
Dautenhahn, K. 2000a. “Socially intelligent agents and the primate social brain: Towards a
science of social minds”. Proc. AAAI Fall Symposium Socially Intelligent Agents — The
Human in the Loop. AAAI Technical Report FS-00-04, AAAI Press, 35–51.
Dautenhahn, K. 2000b. “Design issues on interactive environments for children with
autism”. Proc. The 3rd International Conference on Disability, Virtual Reality and
Associated Technologies. ICDVRAT 2000, Alghero, Italy, 153–161.
Dautenhahn, K. 2002. “The origins of narrative: In search of the transactional format of
narratives in humans and other social animals”. International Journal of Cognition and
Technology 1(1): 97–123.
Dautenhahn, K. 2003. “Roles and functions of robots in human society: Implications from
research in autism therapy”. Robotica 21: 443–452.
<DEST "dau-r27">
"dau-r21">
"dau-r22">
"dau-r23">
"dau-r24">
"dau-r25">
"dau-r26">
Dautenhahn, K. (ed). 2000. Human Cognition and Social Agent Technology. Amsterdam: John
Benjamins.
Dautenhahn, K. and Billard, A. 2002. “Games children with autism can play with robota, a
humanoid robotic doll”. In S. Keates, P. J. Clarkson, P. M. Langdon and P. Robinson
(eds), Universal Access and Assistive Technology. London: Springer-Verlag, 179–190.
Dautenhahn, K. and Werry, I. 2000. “Issues of robot-human interaction dynamics in the
rehabilitation of children with autism”. Proc. From Animals to Animats, The Sixth
International Conference on the Simulation of Adaptive Behavior (SAB2000), Paris,
519–528.
Dautenhahn, K. and Werry, I. 2002. “A Quantitative technique for analysing robot-human
interactions”. Proc. IROS2002, IEEE/RSJ International Conference on Intelligent Robots
and Systems, Lausanne, 1132–1138.
Dautenhahn, K., Werry, I., Rae, J., Dickerson, P., Stribling, P., and Ogden, B. 2002. “Robotic
playmates: Analysing interactive competencies of children with autism playing with a
mobile robot”. In K. Dautenhahn, et al. (2002b), 117–124.
Dautenhahn, K., Werry, I., Salter, T., and te Boekhorst, R. 2003. “Towards adaptive autono-
mous robots in autism therapy: Varieties of interactions”. IEEE International Symposium
on Computational Intelligence in Robotics and Automation (CIRA’03), Kobe, 577–582.
Dautenhahn, K., Bond, A., Cañamero, L., and Edmonds, B. (eds). 2002. Socially Intelligent
Agents — Creating Relationships with Computers and Robots. Dordrecht: Kluwer
Academic Publishers.
Dautenhahn, K. and Nehaniv, C. L. (eds). 2002. Imitation in Animals and Artifacts. Cam-
bridge, MA: The MIT Press.
Dautenhahn, K. and Numaoka, C. (eds). 1998. Socially Intelligent Agents, Part I. Special Issue
of Applied Artificial Intelligence, Volume 12, Numbers 7–8.
Dautenhahn, K. and Numaoka, C. (eds). 1999. Socially Intelligent Agents, Part II. Special
Issue of Applied Artificial Intelligence, Volume 12, Number 3.
Druin, A. and Solomon, C. 1996. Designing Multimedia Environments for Children. Chiches-
ter: Wiley.
Druin, A. and Hendler, J. (eds). 2000. Robots for Kids: Exploring New Technologies for
Learning. San Francisco: Morgan Kaufmann Publishers.
Escalona, A., Field, T., Nadel, J., and Lundy, B. 2002. “Brief report: Imitation effects on
children with autism”. Journal of Autism and Developmental Disorders 32(2): 141–144.
Farnell, B. 1999. “Moving bodies, acting selves”. Annual Review of Anthropology 28: 341–373.
Fogg, B. J. 1999. “Introduction: Persuasive technologies”. Communications of the ACM.
42(5): 27–29.
Frith, U. 1989. Autism: Explaining the Enigma. Oxford: Blackwell.
Frith, U., Morton J., and Leslie, A. 1991. “The cognitive basic of a biological disorder”.
Trends in Neurosciences 14: 433–438.
Frith, U., and Happé, F. 1994. “Autism: beyond ‘theory of mind’”. Cognition 50: 115–132.
Gallagher, S. (1995). “Body schema and intentionality”. In J. L. Bermudez, A. Marcel, and N.
Eilan (eds), The Body and the Self. Cambridge, MA: The MIT Press.
Grandin, T. 1995. Thinking in Pictures. New York: Doubleday.
Grandin, T. and Scariano, M.M. 1996. Emergence: Labeled Autistic. New York: Warner Books.
<DEST "dau-r45">
"dau-r28">
"dau-r29">
"dau-r30">
"dau-r31">
"dau-r32">
"dau-r33">
"dau-r34">
"dau-r35">
"dau-r36">
"dau-r37">
"dau-r38">
"dau-r39">
"dau-r40">
"dau-r41">
"dau-r42">
"dau-r43">
"dau-r44">
Ruffman, T., Garnham, W., and Rideout, P. 2001. “Social understanding in autism: Eye gaze
as a measure of core insights”. Journal of Child Psychology and Psychiatry and Allied
Disciplines 42(8): 1083–1094.
Russell, J. (ed). 1997. Autism as an Executive Disorder. Oxford: Oxford University Press.
Salter, T., te Boekhorst R., and Dautenhahn, K. 2004. “Detecting and analysing children’s
play styles with autonomous mobile robots: A case study comparing observational data
with sensor readings”. Proc. IAS-8, 8th Conference on Intelligent Autonomous Systems,
Amsterdam. IOS Press, 61–70.
Schäfer, S. 1997. “Sterne, Äpfel und rundes Glas: Mein Leben mit Autismus”. Stuttgart:
Verlag Freies Geistesleben und Urachhaus GmbH.
SIA 2000. AAAI Fall Symposium Socially Intelligent Agents — The Human in the Loop, North
Falmouth, MA. Retrieved March 19, 2004, from http://homepages.feis.herts.ac.uk/
~comqkd/SIA-2000.html.
Sloman, A. 1995. “Exploring design space and niche space”. Proc. 5th Scandinavian Confer-
ence on AI. Trondheim, Amsterdam: IOS Press.
Stone, W. L., Ousley, O. Y., Yoder, P. J., Hogan K. L., and Hepburn, S. L. 1997. “Nonverbal
communication in two- and three-year-old children with autism”. Journal of Autism
and Developmental Disorders 27(6): 677–696.
Strickland, D. 1996. “A virtual reality application with autistic children”. Presence: Tele-
operators and Virtual Environments 5(3): 319–329.
Strickland, D., Marcus, L., Hogan, K., Mesibov G., and McAlister, D. 1995. “Using virtual
reality as a learning aid for autistic children”. Proc. of the Autisme 3rd International
Conference on Computers and Autism, Nice, 119–132.
Tapscott, D. 1998. Growing up Digital: The Rise of the Net Generation. New York: McGraw-Hill.
Tardiff, C., Plumet, M. H., Beaudichon, J., Waller, D., Bouvard M., and Leboyer, M. 1995.
“Micro-analysis of social interactions between autistic children and normal adults in
semi-structured play situations”. International Journal of Behavioural Development 18
(4): 727–747.
Tiegerman, E. and Primavera, L. 1981. “Object manipulation: An interactional strategy with
autistic children”. Journal of Autism and Developmental Disorders 11(4): 427–438.
Trevarthen, C. 2001. “Intrinsic motives for companionship in understanding: Their origin,
development, and significance for infant mental health”. Infant Mental Health Journal
22(1–2): 95–131.
Užgiris, I., Benson, J., Kruper, J., and Vasek, M. 1989. “Establishing action-environment
correspondences: Contextual influences on imitative interactions between mothers and
infants”. In J. Lockman and N. Hazen (eds), Action in Social Context, New York:
Plenum, 103–127.
Watson, L. R., Lord, C., Schaffer, B., and Schopler, E. 1989. Teaching Spontaneous Communi-
cation to Autistic and Developmentally Handicapped Children. New York: Irvington
Publishers.
Weir, S. and Emanuel, R. 1976. Using Logo to Catalyse Communication in an Autistic Child.
DAI Research Report No. 15, University of Edinburgh.
Werry, I. 2003. Development and Evaluation of a Mobile Robotic Platform as a Therapy Device
for Children With Autism. PhD thesis, Department of Cybernetics, University of
Reading.
<DEST "dau-r78">
"dau-r77">
Authors’ addresses
Kerstin Dautenhahn, Iain Werry
Adaptive Systems Research Group
Department of Computer Science
University of Hertfordshire,
College Lane
Hatfield, Herts AL10 9AB
United Kingdom
Email: [email protected], [email protected]